AI出错会给企业带来什么后果？

还记得去年的感恩节吗？乔治叔叔坚持要尝试炸火鸡，然后不得不让Alexa（亚马逊人工智能助手）打电话给服务热线Butterball Turkey Talk-Line求助。如果问题复杂，乔治叔叔听起来又很惊慌，接电话的就有可能是人工客服，比如经验丰富的Butterball老员工玛吉·克林德拉。而如果问题简单易答，接电话的就可能是一个复杂的算法系统，它可以识别数百个和火鸡有关的常见词语，并使用玛吉的录音作答。

乔治叔叔有可能不知道（或者不在乎）谁接的电话，前提是他能够快速而顺利地得到了自己想要的准确答案。消费者也是这么想的。《互联网时代：用人工智能带来卓越客户体验》（The Age of Intent: Using Artificial Intelligence to Deliver a Superior Customer Experience）是一本极好的新书，作者P·V·康南（与乔希·贝诺夫合著）在书中写道：“虚拟助手和人类进行合作的最佳商业环节就是帮助客户。机器人迅速而准确，人则感性而且有判断力。结合两者的优势可以打造出完美的客户服务。”

康南是圣何塞咨询与软件公司[24]7.ai首席执行官。在这本书中，他深入研究了几十家公司使用AI来取悦新客户并发展回头客的情况，其中包括租车公司安飞士和旅游服务公司Xanterra。

让机器人或其他虚拟客服代表处理常见问题，较难问题则转给人工客服，这样可以降低成本。卫星电视公司Dish Network称，如果人工客服每分钟能少接一些电话，该公司每年就能节省200万美元。每年顾客会打进约600万个电话，而虚拟客服代表每年都能为Dish Network节省“数千万美元”的客户关系管理预算。

但康南认为，远比这重要的是，用AI来提高销售和服务的速度及精度，并使之更为有趣的公司很快就能建立巨大的竞争优势，这类似于20多年前率先使用互联网的公司。

同时，《世界是平的》（The World Is Flat）一书作者托马斯·弗里德曼在本书前言中写道，智能机器将“改变非机器，也就是人的工作。人们需要更为感性并掌握更详尽的知识……[而且]他们的工作将是调整虚拟客服代表，以便它们变得更聪明。”

《互联网时代》这本书很有趣，其间点缀了一些AI体验出现偏差的警示性事例。它一字不差地复述了另一位作者乔希·贝诺夫和Expedia聊天机器人的一长串对话，内容非常滑稽而且没有任何成效。暴跳如雷的贝诺夫最后打出的一句话是：“你可真傻。”机器的回应则是重复了它已经问了两遍的愚蠢问题。

不过，《互联网时代》的大多数内容都集中在迄今为止哪些奏效及其奏效的原因上。让我们再回到Dish Network，该公司的DiVA系统可以回答常规问题，并且把更复杂的问题留下来，从而帮助人工客服更好、更快地工作。康南指出：“就算DiVA将对话转给客服，它也会在后台继续运行，它会告诉客服出现了什么情况，它认为顾客的问题在哪里以及帮助顾客的最佳途径是什么。”康南还说，这样的合作会让客服人员感到高兴，因为机器去除了工作中乏味的部分，并让他们把自身技能用于“更需要一些关心和共情”的任务上。

世界上其他让AI和人工客服合作的公司都有类似于DiVA的系统。比如，荷兰皇家航空公司的算法已经可以识别约60万次客户电话咨询中的字眼和词汇。系统会持续实时地学习这些互动内容，并筛选海量内部数据，然后为人工客服提供建议，后者则能迅速运用相关知识解答客户疑问。

此类团队协作“需要一个足够聪明的系统来连接公司的所有数据库，包括在周末和假期回答问题的数据库。然后扩大其规模以满足需求，并为人工客服提供支持。”《互联网时代》全面地探讨了其中涉及的复杂因素，不光是建立恰当的基础设施等科技细节，还有让人类成员投入工作所需要的条件。

比如，考虑到最近的那些自动化取代人工的炒作（大多都是夸张，但这是另一个问题），你如何说服人工客服接受甚至欢迎机器提出的建议？关键的一点是要从现在就开始强调智能系统的作用是协助人类，而不是取代人类。

同时，康南建议“谨慎处理内部关系”。很显然，公司的首席技术官和客服（有时是销售）部门负责人必须参与进来。但也别忘了产品开发主管、首席营销官、首席运营官以及人力资源主管，你不可避免地会以这样或那样的方式触及他们的领域。康南写道，厨房里有了如此之多的厨师之后，“就会有许多可能说不的人”。但“除非他们都支持这个决定，否则就不大可能取得进展”。

准备向前迈进的企业或许得注意连锁餐厅TGI Fridays的经验带来的三点启示。千禧一代不像他们的父母那么倾向于在休闲餐厅里吃饭，为了吸引他们，这家公司开发了AI辅助系统，把对社交媒体友好的聊天机器人打造成了明星，此举使其外卖订单在短短一年时间里增长了一倍，达到1.5亿美元左右。

首先，Fridays从自己的一小部分业务着手，具体来说是为了节省时间而预先点菜的顾客，然后在推广这个系统前纠正了其中的所有错误。其次，康南引用该公司负责此项工作的高管谢里夫·米特雅斯的话说：在这个过程中“要彻底衡量所有东西”，以便判断你的系统是否带来了你和顾客想要的结果。

第三，着手这些工作时对上述结果要有清晰的预期。康南写道，TGI Fridays在推广AI方面取得成果的一大原因是它“没有脱离自己的目标。只要选对了要解决的问题，你就开始走向成功了”。（财富中文网）

译者：Charlie

审校：夏林

Remember last Thanksgiving, when your Uncle George insisted on deep-frying the turkey—and then had to ask Alexa to call the Butterball Turkey Talk-Line for help? If the trouble was complicated, and especially if George sounded panicky, he probably reached a real person, like knowledgeable longtime Butterball employee Marge Klindera. On the other hand, with a quick and easy problem, a complex system of algorithms might have connected him instead to a recording of Marge's voice, programmed to recognize and respond to hundreds of common turkey-related phrases.

Chances are, George didn't know (or care) which was speaking to him, as long as he got exactly what he was after, fast and hassle-free. That's what customers want, too. "There's no part of business where virtual agents and humans can work together better than in helping customers," writes P.V. Kannan in a fascinating new book, The Age of Intent: Using Artificial Intelligence to Deliver a Superior Customer Experience (with Josh Bernoff). "Bots are fast and accurate. People are empathetic and have judgment. Together they've got what it takes to deliver the best customer service possible," he writes.

In these pages, Kannan, who is CEO of San Jose-based consulting and software firm [24]7.ai, delves into how dozens of companies, from Avis to Xantera, are using artificial intelligence to delight new customers and keep old ones coming back.

Letting bots or other virtual customer service representatives (CSRs) handle routine queries, while directing trickier challenges to humans, can cut costs. Dish Network says that for every minute, on average, it can shave off customer phone calls with humans, the company can save $2 million per year. Customers initiate about 6 million chats per year, and the virtual CSRs slash "tens of millions of dollars" annually from Dish's CRM budget.

But the far larger point is that companies that deploy A.I. to make sales and service quicker, more accurate, and more fun will soon lay claim to the same kind of huge competitive advantage that early adopters of the Internet enjoyed 20-odd years ago, according to Kannan.

At the same time, smart machines will "change the work of non-machines—also known as 'human beings,'" writes The World Is Flat author Thomas Friedman in this book's foreword. "They will need more empathy and more granular knowledge... [and] their job will be to tweak the virtual agents to make them smarter."

The Age of Intent is fun to read, sprinkled with cautionary tales of A.I. experiments gone awry. A hilarious verbatim transcript of a long, fruitless online dialogue between co-author Josh Bernoff and a chatbot on Expedia ends with an exasperated Bernoff typing, "You're pretty stupid." The bot responds by repeating the same inane question it had already asked twice.

Most of the The Age of Intent, though, focuses on what's working well so far, and why. To go back to Dish Network for a moment, consider how the company's DiVA system, which culls routine queries from thornier ones, helps human customer service agents do their jobs better and faster. "Even after DiVA kicks a chat out to an agent, it continues running in the background," Kannan writes, "where it shares with the agent what already happened, what it believes the customer's problem is, and what might be the best way to help her." The collaboration makes the human reps happier, he adds, because it takes the drudgery out of the job and lets them put their skills to work on the tasks that "require a bit more care and empathy."

DiVA is similar to systems at other companies around the world where A.I. is collaborating with human CSRs. At KLM Royal Dutch Airlines, for instance, algorithms have been programmed to recognize words and phrases from about 60,000 customer chats. Constantly learning from those interactions in real time, while also sifting through oceans of internal data, the system then offers suggestions to the human agent, who can quickly apply the relevant knowledge to the customer's question.

That kind of teamwork "demands a system smart enough to connect to all the company's databases, one that answers questions on weekends and holidays, scales up to meet demand, and supports human agents," Kannan notes. As you'd expect, building that capability is no day at the beach. The Age of Intent takes a thoughtful look at the complexities involved — not only the technical nuts and bolts, like putting the right digital infrastructure in place, but doing what it takes to get the human side humming along too.

How, for example, do you persuade your human CSRs to start taking, and even welcoming, advice from machines? Thanks to all the recent hype about automation taking people's jobs (mostly exaggerated, but that's another story), it's essential to stress from the outset that smart systems are intended to assist human agents, not replace them.

Meanwhile, Kannan suggests, "Navigate politics carefully." It's pretty obvious that a company's CIO and customer-service (and sometimes sales) leaders will have to be on board. But don't overlook the head of product development, the CMO, the COO, and the chief of human resources, all of whose turfs will be trod upon in one way or another. With so many cooks in the kitchen, "there are many potential people who can say no," Kannan writes. But "unless they're all allied behind the decision, it's unlikely to get off the ground."

Businesses ready to forge ahead might bear in mind three insights from restaurant chain TGI Fridays' experience. Aiming to appeal to millennials, who are less inclined to linger in casual-dining eateries than their parents were, the company developed an A.I.-assisted system, starring social-media-friendly chatbots, that doubled its off-premises orders in a single year, to about $150 million annually.

Fridays did this by, first, starting with a relatively small part of its business—people looking to save time by ordering in advance—and working out any bugs before expanding the system. Second, Kannan quotes Sherif Mityas, the Fridays executive who spearheaded the effort: "Measure the hell out of everything" along the way, to see whether your system is producing the results you and your customers want.

And third, go in with a clear idea of what those results would look like. A big reason why TGI Fridays' A.I. push has paid off is that the company "didn't lose track of what they were trying to accomplish," Kannan writes. "Once you've chosen the right problem to solve, you'll be in a position to start succeeding."